Physics contribution
Analysis and reduction of 3D systematic and random setup errors during the simulation and treatment of lung cancer patients with CT-based external beam radiotherapy dose planning

https://doi.org/10.1016/S0360-3016(00)01413-9Get rights and content

Abstract

Purpose: To determine the magnitude of the errors made in (a) the setup of patients with lung cancer on the simulator relative to their intended setup with respect to the planned treatment beams and (b) in the setup of these patients on the treatment unit. To investigate how the systematic component of the latter errors can be reduced with an off-line decision protocol for setup corrections.

Methods and Materials: For 39 patients with CT planning, digitally-reconstructed radiographs (DRRs) were calculated for anterior-posterior and lateral beams. Retrospectively, the position of the visible anatomy relative to the planned isocenter was compared with the corresponding position on the digitized simulator radiographs using contour match software. The setup accuracy at the treatment unit relative to the simulator setup was measured for 40 patients for at least 5 fractions per patient in 2 orthogonal beams with the aid of an electronic portal imaging device (EPID). Setup corrections were applied, based on an off-line decision protocol, with parameters derived from knowledge of the random setup errors in the studied patient group.

Results: The standard deviations (SD) of the simulator setup errors relative to the CT planning setup in the lateral, longitudinal, and anterior-posterior directions were 4.0, 2.8, and 2.5 mm, respectively. The SD of rotations around the anterior-posterior axis was 1.6° and around the left-right axis 1.3°. The setup error at the treatment unit had a small random component in all three directions (1 SD = 2 mm). The systematic components were larger, particularly in the longitudinal direction (1 SD = 3.6 mm), but were reduced with the decision protocol to 1 SD < 2 mm with, on average, 0.6 setup correction per patient.

Conclusion: Setup errors at the simulator, which become systematic errors if the simulation defines the reference setup, were comparable to the systematic setup errors at the treatment unit in case no off-line protocol would have been applied. Hence, the omission of a separate simulation step can reduce systematic errors as efficiently as the application of an off-line correction protocol during treatment. The random errors were sufficiently small to make an off-line protocol feasible.

Introduction

Despite a dose of 65 Gy, the local complete remission rate in patients with non-small cell lung cancer (NSCLC) is only around 20% (1). Unless the irradiated volumes are decreased, radiation dose escalation can result in unacceptable pulmonary and esophageal toxicity, particularly when concurrent chemoradiotherapy is used (2). Volume reduction may sometimes be achieved by omitting elective nodal irradiation 3, 4, but also by enabling the use of small planning margins. Such small margins would also reduce toxicity at currently standard dose levels. In this paper, we investigate the setup inaccuracies of lung cancer patients and discuss the impact on planning margins.

Treatment planning in accordance with the recommendations of the ICRU 50 report requires the definition of a clinical target volume (CTV), which must encompass the gross tumor and subclinical disease and possibly involved lymph nodes. The CTV must be expanded to a planning target volume (PTV) by some geometrical margin. This margin must guarantee adequate coverage of the CTV during treatment and should therefore be based on knowledge of target movement with respect to the treatment fields, and machine accuracy (e.g., reproducibility of the block positions). However, the ICRU 50 report does not give a clear recommendation on how the CTV-to-PTV margin should be chosen.

Stroom et al. (5) have derived a general calculation method based on the dose coverage probability of the CTV to derive a 3D margin from known setup errors and internal organ movements, which may have any probability density function. They found that a clear distinction must be made between systematic and random errors, where a systematic error is the average setup error of the target volume over all fractions for a certain patient and the random errors are the inter-fraction variations. For gaussian error distributions, with a standard deviation (SD) Σ for the systematic errors and σ for the average random error, a CTV-to-PTV margin of 2Σ + 0.7σ seemed appropriate to guarantee that, on average, 99% of the CTV receives at least 95% of the prescribed dose (if the 95% isodose contour encompasses the PTV in the treatment planning). Their calculations incorporated the presence of small rotations (1 SD ∼ 1°). A very similar result was found by Van Herk et al. (6), who did an analytical calculation for the simplified situation of a spherical target volume in an ideally conformal homogeneous dose. Both of these results confirm the intuitive notion that systematic errors are more important than random errors in establishing planning margins.

Most setup studies in lung cancer patients are 2D 7, 8, 9, 10 or do not properly separate and quantify setup errors in terms of random and systematic components 11, 12, 13. These studies usually pertain to small (< 20) numbers of patients 8, 9, 10, 11, 12. In addition, none of these studies takes into account the setup errors that are made on the simulator [which have been shown to be of importance for other tumor sites 14, 15]. Lastly, the impact of a 3D off-line correction protocol on the systematic errors during treatment has not yet been studied in this patient group. As stressed above, accurate knowledge of both the systematic and the random setup inaccuracies (and how they can be reduced) is a prerequisite for calculation of optimal planning margins. Therefore, we measured the magnitude of both simulator and treatment systematic setup errors as well as random treatment setup errors in 3D in 40 lung cancer patients. In addition, we report on the reduction of systematic treatment errors through an off-line setup verification protocol.

Geographical tumor misses are caused both by external setup errors and by internal movement of the target volume 9, 16. In this study, we focus on the external setup errors for a group of lung cancer patients with CT planning, consisting of: (a) the errors made by using simulator films as definition of the reference setup 14, 15, and (b) the setup errors at the treatment unit relative to the reference setup.

The position of the patient anatomy in the reference setup (during which the treatment isocenter is marked on the patient and reference images are obtained) relative to the isocenter should be in agreement with the corresponding position in the CT treatment plan. However, in many institutions (including, until recently, our own) it is customary to mark the final beam setup at the simulator, after the planning has been performed. The definition of the final isocenter is based on visual inspection, and therefore may deviate from the intended CT plan isocenter. This simulator setup error results in a systematic error in the patient treatment. For prostate (14) and head and neck (15) irradiation, it was found that the simulator setup errors are comparable to systematic setup errors at the treatment unit. In these two cases, bony structures relevant to patient setup can be clearly identified, somewhat in contrast to the thorax region. Therefore, we expect that the simulator setup errors for lung cancer patients will be at least equal to, or larger than, systematic treatment unit setup errors. The simulator setup errors are, however, not well appreciated in the literature on setup accuracy of these patients. In fact, of the seven setup studies we refer to above 7, 8, 9, 10, 11, 12, 13, six defined explicitly which type of setup reference image was used. Of these six studies, five used simulator images based on a classical simulation 8, 9, 10, 11, 12 (i.e., without registering with the digitally-reconstructed radiographs [DRRs]) and one study (13) applied both simulator films and DRRs.

The setup errors at the treatment unit relative to the reference setup have a systematic and a random component. The systematic component may be reduced with portal imaging and setup corrections based on an off-line decision protocol 17, 18. We have investigated both components and report the results of an off-line decision protocol.

Section snippets

Description of setup errors

We adopt the definitions introduced by Bijhold et al. (19) for systematic and random errors. The systematic setup error of a patient is the setup error averaged over all dose fractions. The group mean of this error is denoted by μ and the SD by Σ (i.e., the interpatient mean setup variation). The random error of a patient is the standard deviation (SD) of the setup error from fraction to fraction (i.e., the interfraction variation). An appropriate average is taken to obtain the average random

Intrafraction movement in the Z-direction

For the 8 patients in the pilot study, the random setup error in the Z-direction, as measured from the position of sternum, vertebrae, and trachea, was σZT = 2.2 mm. This random error is as small as the random errors in the X-and Y-direction in a similar patient group (10), which we obtained with structures that were proven to exhibit little intrafraction movement. It is also equal to the random setup error we obtained for over 700 prostate cancer patients (σZT = 2.0 mm) at the same treatment

Discussion

The purpose of investigating the possible displacements of a target volume during treatment relative to the planned situation is to define appropriate planning margins, as well as to identify efficient methods to ensure or even reduce those margins. These margins should account for both internal target volume displacements and external setup errors.

The internal displacement for thoracic tumors is mainly due to breathing and cardiac activity. An estimate of the internal tumor displacements can

Conclusion

We have investigated the systematic and random external setup errors during the irradiation of lung cancer patients. The magnitude of the systematic errors at the simulator and the treatment unit is comparable if no protocol is used to reduce the latter errors Table 1, Table 2, indicating that reduction of the former is no less important than reduction of the latter. Therefore, an off-line treatment correction protocol based on simulator films becomes inefficient (compare methods 2 and 3 in

Acknowledgements

The authors thank Erik van Dieren for providing DRR generation software, the technicians at the MM50 Racetrack Microtron for their continuing enthusiasm in performing the portal image data analysis and setup corrections, and Marcel Eggen for providing data on the treatment table accuracy of the MM50.

References (28)

Cited by (103)

  • Spatial precision in magnetic resonance imaging-guided radiation therapy: The role of geometric distortion

    2016, International Journal of Radiation Oncology Biology Physics
    Citation Excerpt :

    The quadrature sum of the requisite margins resulting from each source of uncertainty can be used as a guide in defining the planning target volume (117). Errors can be classified as either systematic or random (118), and the magnitude of each error component ultimately determines the margin size chosen. Monte Carlo simulations have shown that the margin Δ, measured in millimeters, can be approximated as 2.5Σ + 0.7σ, where Σ is systematic error and σ is random error (119).

  • Intra thoracic anatomical changes in lung cancer patients during the course of radiotherapy

    2014, Radiotherapy and Oncology
    Citation Excerpt :

    We observed ITACs in 72% (N = 128) of all lung cancer patients during the radical radiotherapy course. Regression during treatment was described by several groups [2–14,18]. Knap et al. [5] reported that 1/3 of lung cancer patients undergoing (chemo-) radiotherapy achieved significant tumor shrinkage at the end of radiotherapy.

View all citing articles on Scopus

This work was financially supported by the Dutch Cancer Society (grant DDHK 96-1258).

View full text